Restaurant Review System Using Hidden Markov Model

A restaurant review system has been developed which calculates the ratings of the restaurant by implementing forward and Viterbi algorithm of Hidden Markov Model (HMM). The words of each review are considered as the observable sequences and the tagged levels of those words as hidden states. By using the sequences and states, the Python program calculates the transition and emission probability of HMM model as well as how often the Viterbi/forward algorithm tags a word as positive, negative or neutral. By computing a score (-1 if all words negative, +1 if all words positive), this software calculates the average rating and maps it to a range of 0 to 5.

""" demonstration of the Viterbi algorithm """ import numpy as np class HmmStates: def __init__(self, name, adjstates, emissionprob): self.name = name # name of state self.adjs = adjstates # dictionary of adjacent states + transition probability self.ep = emissionprob # dictionary of emission probabilities class VState: def __init__(self, hmmstate, prob): self.state = hmmstate self.p = prob class Hmm(HmmStates): def __init__(self): """         create HMM from data set         genertated transition and emission lists         """ ## TODO: create the actual HMM self.hmm = [HmmStates('S', {'S': 0.8, 'R': 0.2}, {'H': 0.8, 'G': 0.2}),             HmmStates('R', {'S': 0.4, 'R': 0.6}, {'H': 0.4, 'G': 0.6})] # dummy HMM for testing purpose # prior probabilities TODO: compute prior probabilities from HMM self.prior = {'S': 2/3, 'R': 1/3} def viterbi(self, observation): # compute prob of hidden states for first observed state         ps_old = {} # most likely probabilities of current hidden state         vseq = [None]*len(observation) # most likely sequence         ps_max = 0 for s in self.hmm:             ps_old[s] = self.prior[s.name] * s.ep[observation[0]] if ps_old[s] > ps_max:                 vseq[0] = VState(s, ps_old[s])                 ps_max = ps_old[s]         i = 1 for o in observation[1:len(observation)]:             ps_new = {}             ps_max = 0 for s in self.hmm:                 temp = 0 for sp in ps_old: if ps_old[sp] * sp.adjs[s.name] * s.ep[o] > temp: temp = ps_old[sp] * sp.adjs[s.name] * s.ep[o] #TODO fix this!!!!                 ps_new[s] = temp if temp > ps_max:                     vseq[i] = VState(s, temp) # save state with highest probability                     ps_max = temp             ps_old = ps_new             i += 1 return vseq hmm = Hmm() viterby_seq = hmm.viterbi(['H', 'H', 'G', 'G', 'G', 'H']) for i in range(len(viterby_seq)): print(viterby_seq[i].state.name + ' ' + str(viterby_seq[i].p)) #print(viterby_seq)